Environmental Variables

We selected environmental features (i.e. landscape/land use) that might potentially affect population dynamics of H. halys in New Jersey [5], [48], [53]. These features are classified into five types of landscape/land cover features, which include agriculture, urbanization, wetlands, forests, and distance to nearby roads and highways. Landscape/land use features within the Pinelands and along the Atlantic coast were excluded in the analysis because no traps were placed in these habitats. These five main types of features were obtained by acquiring GIS digital files from the New Jersey Department of Environmental Protection (NJDEP) GIS Land use/Land cover dataset from 2007 [34], [9] (see Figure 1). These files were selected because they represent the most current datasets produced by NJDEP. These data were taken from aerial photography captured in the spring of 2007.

We merged datasets from every county within New Jersey to create individual maps representing the five main types of landscape/land use features using the merger function in ArcMap 10. Trap locations were plotted on these maps; buffers of 2 km were placed around each of the traps, representing the total hectares H. halys may inhabit [52]; and square acreage of 48 land use/land cover variables within the main landscape/land use features (see Table S1 for list of variables and acreage) was calculated using the calculate geometry function in ArcMap 10. Lastly distance, in meters, from each trap to the nearest highways was calculated in ArcMap 10. Square hectares of all 48 variables, distances to nearest highway, and total population of H. halys for each trap were entered into Excel for statistical analysis.

Kernel Density Estimation

Kernel Density Estimation (KDE) was implemented to examine the overall invasion dynamics and colonization of H. halys using the geographical information systems software ArcMap 10. KDE uses probability statistics (i.e. kernel function) to calculate the estimated density of a population around the actual population data points producing a fitted, smooth, tapered surface that connects all data points used in the analysis [9], [42]. We selected KDE because this analysis performs well with small amounts of data, they are robust to autocorrelation, and results are in a utilization distribution (i.e. UD; A UD is a grid where the value for each cell represents the probability of the taxa occurring in that cell) rather than a simple distribution range outline [21].

KDEs were derived from H. halys density collected from 2004 through 2011 and bandwidths (i.e. smoothing parameter). In calculating these bandwidths, we randomly measured distances between 20 traps, and then averaged those distances. A bandwidth specifies the maximum distance at which data points (i.e. traps) are distributed over the study area or the search radius of the Kernel Density function [9].

Hot-Spot Analysis

Hot-Spot Analysis was used to detect statistically significant populations of H. halys over New Jersey from 2004 to 2011. This was done to provide information on the overall dispersal and distribution of H. halys throughout New Jersey and provide insight on the relative rate of spread of H. halys populations across the New Jersey landscape. This analysis was used to compliment KDE because it can detect statistically significant (i.e., Z-scores) high (i.e., hot spot) and/or low (cold spot) clustering of H. halys populations. To be a statistically significant hot spot, a farm with a high density of H. halys has to be surrounded by neighboring farms with high densities of H. halys. The local sum of H. halys in a farm and its neighbors is then compared to the sum of all farms for each year; when the local sum is very different from the expected local sum, and that difference is too large to be the result of random chance, a statistically significant Z-score will result [9]. Z-scores are standard deviations, associated with p-values found in the tails of the normal distribution.

Bayesian Analysis

Bayesian linear regression was implemented to examine which of the 48 landscape/land use factors are facilitating the invasion, establishment, and dispersal of H. halys populations in New Jersey from 2004 to 2011 using the statistical software WinBUGS (Bayesian inference Using Gibbs Sampling; [43]). Bayesian linear regression was selected over univariate [e.g. analysis of variance (ANOVA)] and multivariate [e.g. principal component analysis (PCA)] statistical methods because it is a powerful and robust methodology that allows the observer to combine total H. halys population data with additional, independently available information (e.g. landscape/land use variables; the prior) to produce a full probability distribution (posterior distribution) of the relationships between H. halys populations and various landscape/land use factors; allows one to make inferences based on small sample sizes; and it appropriately models uncertainty on those relationships [6], [8], [10], [11], [23]. Moreover, the Bayesian methodology allows us to ask directly how probable is the hypothesis (e.g. Ho: acreage of different agriculture and urban landscape/land use factors will influence H. halys populations over time and space), given the data; whereas, conventional statistical methods lack this advantage. The following Bayesian regression model (Equation 1) was fitted to the hectare and total H. halys data, using reversible jump MCMC methods:

Equation 1. Bayesian Regression Model used in the analysis.

Reversible jump MCMC (RJMCMC) was selected over other Markov chain algorithms because this algorithm allows for sampling of posterior distributions that are generated from an unequal numbers of traps (i.e., samples) [13], [14], [29]. RJMCMC computations are generated by computing and converging on the most optimal posterior distributions. These computations are accomplished by fitting complex models (e.g. linear regression model) to our H. halys population and landscape/land use data. For relationships between H. halys and landscape/land use data to be considered informative, the posterior distribution of that relationship must be 0.05 or greater [i.e., a high posterior probability (i.e., α-level = 0.05), which indicates a statistically significant relationship between total H. halys population and a particular environmental variable]. RJMCMC sampled from 1,000,000 iterations and variable selection, which reflected a strong association (i.e., high probability) with the H. halys data, was based on the final 500,000 iterations [29]–[31].

The notations of σ² and β in the model above (Equation 1) represent the prior distributions on our parameters, where IG (inverse-gamma distribution) represents the marginal posterior distribution for the unknown variance of a normal distribution when an uninformative prior is used. These priors were chosen to be non-informative or vague because we are primarily interested in understanding what relationships are informative between landscape/land use variables and total H. halys populations [23]. Other notations in the model include the dependent variables (i.e., Y₁, Y₂,…,Y₈). More specifically, Y_n represents the total H. halys populations collected from each farm from 2004 through 2011. Lastly, X represents the independent variables or covariates (i.e., 48 landscape/land use variables).

Environmental Factors

The Bayesian linear regression identified 41 out of the total 48 landscape/land use variables that displayed a statistically significant relationship with densities of H. halys from 2004 through 2011 (Table 1). More specifically, in 2004 we observed intensive land use, for example, communications and utility buildings, suburban communities, and parking lots, displayed a positive relationship with densities of H. halys. As the populations dispersed east and south in 2005 (Figure 2B) we observed strong associations with not only intensive land use, but also with commercial land use (e.g. strip malls) and railroads (Table 1, Figure 4). As H. halys populations in 2006 increased in density and continued to disperse east and south (Figure 2C), we found this insect displayed strong associations with urban factors, such as railroads, and other landscape/land use features that include deciduous forests and wetlands (e.g. fallow agricultural land and wetland communities either natural or planted that are maintained in residential, commercial, and or industrial areas; Figure 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Summary table of landscape/land use variables that displayed significant positive relationship with densities of Halyomorpha halys using Bayesian linear regression analysis.

Pink represents urban landscape/land use variables; green represents forest landscape/land use variables; blue represents wetland landscape/land use variables; and red represents agricultural landscape/land use variables. In 2010 and 2011 we calculated the total number of these relationships observed.

https://doi.org/10.1371/journal.pone.0095691.g004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Probability distributions calculated from total yearly densities of Halyomorpha halys captured from black light traps placed on farms from 2004 to 2011 using Bayesian linear regression and the reversible jump MCMC algorithm.

https://doi.org/10.1371/journal.pone.0095691.t001

In 2007 as H. halys populations continued to move further south they displayed strong positive relationships with commercial structures and wetlands rights-of-way (former wetlands converted to road with some shrub vegetation; Figure 4). By 2008 and 2009, this insect dispersed further south (Figures 2E–F) and showed positive relationships with coniferous forests, and wetland landscapes (e.g., wetland right-of-way, Phragmites, fallow agriculture land, and mixed wetlands), and less so with urban factors, for example intensive urban land use and storm water basins (i.e., surface water collection site, associated with new commercial and residential areas), compared to 2005 and 2006 (Table 1, Figure 4). Lastly, from 2010 and 2011, H. halys displayed positive significant relationships with the majority (63–92%) of New Jersey landscape features examined. Among these landscape/land use features, agricultural variables, for example vineyards and orchards (Table 1, Figure 4) displayed positive relationships with this insect. Moreover, from 2010 to 2011 we observed a 70% increase in relationships between agriculture variables and H. halys. Although, we observed a trend in associations between H. halys and various landscape/land use features, relationships between this insect pest and distance to nearest highways and major roads from 2004 through 2011 were absent (Figure 4).

First Phase: Initial Establishment and Dispersal

Within the first two years, from 2004 to 2005, the initial dispersal of H. halys began in the northwestern portion of New Jersey (Warren and Hunterdon Counties; Figures 2A). These populations were in close proximity to where this insect pest was first detected in Allentown, PA to the east (Somerset and Mercer Counties). A small population in the south (Burlington and Camden Counties) was later detected in 2005 (Figure 2B). Accompanying this initial dispersal we observed an increase in density from 2004 to 2005 (Table S1). Landscape/land use factors displaying positive relationships with this dispersion include: intensive urban land use, commercial developments (e.g., residential houses, strip malls, and supermarkets), and railroads (Figure 4). These factors may have been instrumental during the initial dispersion and establishment because these commercial and residential developments have provided adequate protection and increased the overwintering survival of H. halys [22], [19].

Another possible explanation for the patterns observed from 2004 to 2005 is the Allee effect. The Allee effect is a positive relationship between the number of individuals in a population and their fitness [1]. Meaning that an individual of a species that is subject to an Allee effect will suffer a decrease in some aspect of its fitness when population density is low [1], [44], [45], [3]. The population of H. halys is hypothesized to be severely, genetically bottlenecked and arising from a single introduction with a small propagule size [51] as small as two females. Theoretically, this would cause a significant reduction in fitness caused by inbreeding depression through Allee effect. However, as evidenced by our results, H. halys appears to have successfully overcome the Allee effect and spread at a rate of 2.84 farms per year with a 75% annual population increase [38]. The correlation between urban and residential developments during this establishment phase may have reduced the Allee effect of H. halys, thus facilitating a higher rate of establishment and dispersion. A similar pattern was observed in French Polynesia for the glassy-winged sharpshooter, Homalodisca vitrepinnis (Germar). For example, islands experiencing the early stages of invasion by H. vitrepinnis displayed high densities in urbanized areas with substantially lower abundances on remote islands further away from the main population [39].

The small satellite population of H. halys observed in Burlington and Camden Counties may also be a result of the Allee effect (Figure 2B) as has been shown in other insects [39]. Brown et al. [4] observed that many species of passerine birds displaying a reduced Allee effect have relatively clustered distribution with many sites of low abundance and few hot-spots of high local abundance.

Human-transportation activity (i.e., the movement of plants and contaminated cargo via ships, cars, boats, and rails) may also explain the increase in H. halys density and its establishment and range expansion in northwestern New Jersey. These activities can deliberately move material while simultaneously and unintentionally moving pest species, such as H. halys [12], [39]. Of these activities, we observed a positive relationship between H. halys density and railroads in 2004 and 2005 (Figure 4). This suggests that railroads, also present in northwestern New Jersey (Warren and Hunterdon Counties), may have eased barriers to establishment resulting in the movement of this insect into new habitats. The association with railroads has been demonstrated in the establishment and dispersion of other insect pests, such as the glassy-winged sharpshooter in French Polynesia [39] and the rice water weevil in China [15]; although additional monitoring and analysis is required to adequately elucidate the relationship between H. halys and railroads.

In addition, these railroads and urban/residential developments may be surrounded by high host-plant diversity for H. halys. Two such hosts are Paulownia tomentosa (Thunb.) Sieb. and Zucc. ex Steud. (i.e., princess tree) and Ailanthus altissima (Mill.) Swengel (i.e., tree of heaven), natives to China [7], [35], [25]. These tree species are ornamentals, found in parks, gardens, and can be found in highly disturbed environments (e.g. intensive urban development and potentially near railroads). Because of the potential close proximity of these trees to railroads and urban/residential developments the following effect may have been produced: P. tomentosa and A. altissima are acting as a sink for H. halys, which could accelerate their invasion into urban/residential developments thereby increasing its abundance and reducing the Allee effect; and H. halys could readily move from P. tomentosa and A. altissima to nearby rail cars.

Lastly, orchard trees may be facilitating the movement of this pest insect to rail cars. These orchard trees are another potential source of establishment and dispersion because they are known hosts of H. halys [26]. Moreover, these orchards occupy approximately 434 hectares in Warren and Hunterdon Counties [49], which were the first counties to harbor this pest insect. Despite this evidence, additional investigations are needed to examine the associations between orchards, railroads, and H. halys in early establishment.

Second Phase: Population Growth and Range Expansion

After the initial establishment and dispersal of H. halys in 2004 and 2005, we observed rapid population growth and range expansion from 2006 to 2008 (Figure 2D–F). This growth and range expansion was associated with an increase in the insect’s relationship with urban, forest, and wetland landscape/land use variables than the previous years (Table 1, Figure 4). The agriculture/urban interface, which is common in New Jersey, may have facilitated population growth of H. halys by offering agricultural and cultivated host plants for development and natural and human-made overwintering structures. The association with human habitat may be specifically important and has been suggested to increase overwintering survival relative to pentatomid species that overwinter in natural habitats [22]. As a consequence, we hypothesize that populations of H. halys may have rapidly increased, overcoming a critical minimum density threshold, followed by the establishment of new populations observed in eastern and southern New Jersey (Figures 2C–E, 3C–E), and thus reducing a possible Allee effect.

In addition to overcoming the Allee effect, the long distance movement from northwestern New Jersey to the southernmost portion of the state from 2006 to 2008 (Figures 2C–E) may be a consequence of stratified diffusion. Stratified diffusion refers to the movement of an invasive species by rapid long-distance transport and local diffusion around a point of infestation [39]. Long-distance movement may have been facilitated by rail cars to central New Jersey in 2006 (Figure 4). Moreover, these railroads are located in areas where high densities of H. halys have been observed in 2006 (Figure 2C). In 2007 and 2008 long distance-movement from central New Jersey to the southernmost part of the state may be attributed to close proximity of commercial developments (e.g., strip malls, motels, and gas stations) acting as stepping-stones for invasion and fruit tree orchards or other adjacent suitable habitats near wetland rights-of-way (Figure 4). It should be noted that no relationships were detected between H. halys and highways/roads because black lights may have been too far away from this pathway.

The second component of stratified dispersion is local diffusion which occurs when a nascent colony becomes established and begins to expand naturally. This diffusion of H. halys after the long distance movement, appearing in 2007 (3 years after initial establishment to the southernmost region of New Jersey), may have been facilitated by encountering new habitats for overwintering, such as deciduous forests [i.e., dead, standing trees (D–H. Lee, unpublished data)]. Other factors facilitating this diffusion may include suitable host plants near fallow agriculture that has undergone succession into wetlands, and the increased hectares of orchard trees located in the southernmost region of New Jersey, such as Gloucester and Burlington Counties [49].

Phase 3: Potential Spurious Lag Phase

Following this population growth and expansion phase observed from 2006 to 2008 (Figures 2C–E), was a subsequent reduction in population size of H. halys in 2009 (Figure 2F, Table S1). This reduction coincides with a decrease in the number of positive significant relationships observed between H. halys and landscape/land use variables (Table 1, Figure 4). More specifically, only wetland variables, such as herbaceous wetlands, mixed shrub/scrub deciduous and coniferous wetlands, and Phragmites wetlands (Table 1, Figure 4) displayed significant positive relationships with this insect. These results suggest these wetland habitats could be barriers for mate finding of H. halys [45], although additional data are needed to examine the impact of these wetlands on the reproduction of this insect. Another more probable explanation in the decrease of H. halys observed in 2009 may be a sampling artifact, caused by 51 black light traps used in 2009 compared to 71 black light traps used in 2008. This potential spurious result is further supported in Nielsen et. al [38] study, which showed a continuing increase in populations of H. halys from 2004–2011 in New Jersey when black light traps are kept constant.

Phase 4: Exponential Growth

An exponential growth phase was observed from 2010 to 2011 (Table S1), as well as H. halys possibly dispersing and/or populations increasing from the north to the south (Figures 1G–H, 2G–H). This pattern was associated with all the main landscape/land use categories, with the exception of major highways and roads (Table 1, Figure 4). Positive density increases of H. halys and their subsequent expansion into the southernmost part of the state may be attributed to higher acreage of suitable habitats, in particular orchards. This was supported by a positive relationship observed between orchards and H. halys (Table 1). The high acreage of orchard trees in southern New Jersey [49], which are important hosts of this insect [25], [26], [36], may have supported high densities of H. halys. Although, continued monitoring in and near these orchards is needed to increase confidence in this trend.

Although there are many positive aspects of our dataset, our conclusions are dependent upon the location of the black light traps themselves. In this study, the black light traps are always placed in a vegetable farm and thus we have minimal data in non-agricultural areas such as densely populated urban areas in northeastern New Jersey, the Pinelands, and coastal regions. Additional black light traps in these habitats are needed to provide a more complete picture of the invasion and colonization dynamics of H. halys in New Jersey and other regions with diverse landscape/land use features.